Abstract:Partially observable Markov decision processes (POMDPs) provide a modeling framework for a variety of sequential decision making under uncertainty scenarios in artificial intelligence (AI). Since the states are not directly observable in a POMDP, decision making has to be performed based on the output of a Bayesian filter (continuous beliefs). Hence, POMDPs are often computationally intractable to solve exactly and researchers resort to approximate methods often using discretizations of the continuous belief space. These approximate solutions are, however, prone to discretization errors, which has made POMDPs ineffective in applications, wherein guarantees for safety, optimality, or performance are required. To overcome the complexity challenge of POMDPs, we apply notions from control theory. The goal is to determine the reachable belief space of a POMDP, that is, the set of all possible evolutions given an initial belief distribution over the states and a set of actions and observations. We begin by casting the problem of analyzing a POMDP into analyzing the behavior of a discrete-time switched system. For estimating the reachable belief space, we find over-approximations in terms of sub-level sets of Lyapunov functions. Furthermore, in order to verify safety and optimality requirements of a given POMDP, we formulate a barrier certificate theorem, wherein we show that if there exists a barrier certificate satisfying a set of inequalities along with the belief update equation of the POMDP, the safety and optimality properties are guaranteed to hold. In both cases, we show how the calculations can be decomposed into smaller problems that can be solved in parallel. The conditions we formulate can be computationally implemented as a set of sum-of-squares programs. We illustrate the applicability of our method by addressing two problems in active ad scheduling and machine teaching.

Probabilistic Alternating-Time -Calculus.

Probabilistic Alternating-Time <em>µ</em>-Calculus

Probabilistic Alternating-Time mu-Calculus

Reasoning about Strategic Abilities in Stochastic Multi-agent Systems

The Alternating-Time μ-Calculus With Disjunctive Explicit Strategies

Probabilistic Alternating-time Temporal Logic and Model Checking Algorithm

Characterising Probabilistic Alternating Simulation for Concurrent Games

Global Model Checking on Pushdown Multi-Agent Systems

Natural Strategic Ability in Stochastic Multi-Agent Systems

Bayesian Statistical Model Checking for Multi-agent Systems using HyperPCTL*

Monitoring Ctmcs By Multi-Clock Timed Automata

Making Agents' Abilities Explicit

Multi-Valued Verification of Strategic Ability

Responsibility-aware Strategic Reasoning in Probabilistic Multi-Agent Systems

EPMC Gets Knowledge in Multi-agent Systems

UPPAAL-SMC: Statistical Model Checking for Priced Timed Automata

Unknown Biases and Timing Constraints in Timed Automata

What Are the Odds? Improving the foundations of Statistical Model Checking

Statistically Model Checking PCTL Specifications on Markov Decision Processes via Reinforcement Learning

Control Theory Meets POMDPs: A Hybrid Systems Approach

Model Checking Probabilistic Epistemic Logic for Probabilistic Multiagent Systems